Coastal Ocean Forecasting: system integration andevaluation
V.H. Kourafalou, P. De Mey, M. Le Hénaff, G. Charria, C.A. Edwards, R. He, M.Herzfeld, A. Pascual, E.V. Stanev, J. Tintoré, N. Usui, A.J. van der Westhuysen,J. Wilkin & X. Zhu
To cite this article: V.H. Kourafalou, P. De Mey, M. Le Hénaff, G. Charria, C.A. Edwards, R. He, M.Herzfeld, A. Pascual, E.V. Stanev, J. Tintoré, N. Usui, A.J. van der Westhuysen, J. Wilkin & X. Zhu(2015) Coastal Ocean Forecasting: system integration and evaluation, Journal of OperationalOceanography, 8:sup1, s127-s146, DOI: 10.1080/1755876X.2015.1022336
To link to this article: http://dx.doi.org/10.1080/1755876X.2015.1022336
Coastal Ocean Forecasting: system integration and evaluation
V.H. Kourafaloua*,n, P. De Meyb,n, M. Le Hénaffa, G. Charriac, C.A. Edwardsd, R. Hee,n, M. Herzfeldf,n, A. Pascualg,E.V. Stanevh,n, J. Tintorég,n,i, N. Usuij, A.J. van der Westhuysenk,n, J. Wilkinl and X. Zhum,n
aUniversity of Miami, Rosenstiel School of Marine and Atmospheric Sciences, Miami, FL, USA; bLEGOS, Laboratoire d’Etudes enGéophysique et Océanographie Spatiales, Toulouse, France; cIFREMER/DYNECO, Plouzané, France; dUniversity of California, Dept. ofOcean Sciences, Santa Cruz, CA, USA; eNorth Carolina State University, Dept. of Marine, Earth & Atmospheric Sciences, NC, USA;fCSIRO, Hobart, Australia; gIMEDEA CSIC-UIB, Mallorca, Spain; hHelmholtz-Zentrum, Institute for Coastal Research, Geesthacht,Germany; iICTS SOCIB, Palma de Mallorca, Spain; jMeteorological Research Institute, Oceanography and Geochemistry Research Dept.,Tsukuba, Japan; kIMSG at NOAA/NWS/NCEP/Environmental Modeling Center, College Park, MD, USA; lRutgers University, Institute ofMarine and Coastal Sciences, New Brunswick, NJ, USA; mNational Marine Environmental Forecasting Center, Beijing, China; nCoastaland Shelf Seas Task Team Member, GODAE OceanView
Recent advances in Coastal Ocean Forecasting Systems (COFS) are discussed. Emphasis is given to the integration of theobservational and modeling components, each developed in the context of monitoring and forecasting in the coastal seas.These integrated systems must be linked to larger scale systems toward seamless data sets, nowcasts and forecasts (fromthe global ocean, through the continental shelf and to the nearshore regions). Emerging capabilities include: methods tooptimize coastal/regional observational networks; and probabilistic approaches to address both science and applicationsrelated to COFS. International collaboration is essential to exchange best practices, achieve common frameworks andestablish standards.
Introduction
Within GODAE OceanView [GOV; http://godae-oceanview.org], the international Coastal Ocean andShelf Seas Task Team (COSS-TT) aims to consolidatethe foundation and support the advancement of coastalocean forecasting science, systems, and applications.The main goal and central mission of the COSS-TT isto work within GOV, and in coordination with theGlobal Ocean Observing System (GOOS), towards theprovision of a sound scientific basis for sustainable multi-disciplinary downscaling and forecasting activities in theworld’s coastal oceans. The initiative is built aroundthree key concepts: ‘international’, ‘scientific’, and ‘sus-tainable’ and is driven both by science and through thepromotion of good practices. These drivers emerge andadvance through the international coordination of abroad range of scientific approaches and applicationsexamined within individual Coastal Ocean ForecastingSystems (COFS).
The COSS-TT has initiated the consolidation of a broadcoastal scientific community around the main disciplines ofphysics and interactions between physical and biogeo-chemical processes. The strategic goal is to help achievea truly seamless framework from the global to thecoastal/littoral scale. Coastal ocean is defined inclusive
of nearshore and shelf regions, and of the adjacent deepocean part that triggers or is influenced by shelf and shelfbreak processes (Robinson et al. 2004). It is recognizedthat the influence of coastal ocean processes is felt farbeyond the shelf break, thus interacting with open oceandynamics and controlling the connectivity of remote eco-systems. The innovative approach that this internationalactivity advocates and oversees is that forecasting in thecoastal and shelf seas must fully address land-sea, air-sea,and coastal-offshore interactions.
The goal of this paper is to showcase methodologiesintegrating observations and models in coastal areas, insynergy with larger scale observatories and modelingsystems, in support of coastal ocean forecasting. The nextsection discusses recent advances and future challenges incoastal observational networks and models, employingexamples of integrated systems over diverse coastalenvironments around the world and discussing methodsto optimize array design. Then, emerging statisticalapproaches are introduced that also deal with forecastuncertainty. Conclusions synthesize international initiativesand future strategies. A more detailed discussion onspecific scientific topics in support of coastal ocean fore-casting and on COFS applications can be found in a com-panion paper (Kourafalou et al. 2015).
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/Licenses/by/4.0/), which permits unrestricteduse, distribution, and reproduction in any medium, provided the original work is properly cited.
Journal of Operational Oceanography, 2015Vol. 8, No. S1, s127–s146, http://dx.doi.org/10.1080/1755876X.2015.1022336
Integrated coastal ocean forecasting systems (COFS)
COFS combine comprehensive observational networks andappropriate modeling systems to ensure the continuousmonitoring of changes in the coastal ocean and supportforecasting activities that can deliver useful and reliableocean services. Oceanographic information, integratedwith predictive models, is increasingly needed to: sustain-ably manage coastal and ocean areas; portray the oceanstate today, next week and for the next decade; increaseshipping efficiency; mitigate storm damage and floodingof coastal areas; sustain fisheries and fish stocks; protectimportant ecosystems from degradation; help decision-making in times of crisis; and improve climate forecastingin response to global change, among other directapplications.
Coastal ocean monitoring
To achieve reliable model simulations and predictions,COFS require quality-controlled data, either archived orreal-time, on a routine basis. The data are used to: identifythe important processes in the study area set-up the appro-priate numerical models validate model simulations andassess their quality optionally carry out data assimilation,toward enhanced predictive skill or to perform reanalyses.
Although such needs are similar for large-scale oceanforecasting systems, there are distinct differences betweencoastal and global monitoring.
Specific aspects of coastal-ocean monitoring
There are several examples of how large-scale monitoringsystems might not provide the types and/or attributes ofdata needed for COFS. For instance, present-day nadir sat-ellite altimetry, instrumental in ocean forecasting, provid-ing real-time global coverage, does not fully resolve allimportant coastal-ocean scales. Data from profiling floatsare often not available in the shelf seas. Given the compara-tively shorter space/time scales in coastal regions, thequasi-homogeneous sampling characteristics of open-ocean networks are often inadequate.
Coastal regions present the advantage that permanent,multivariate instrumented sites are easier to set up.Coastal observatories can thus employ various datasources. Regionally deployed technologies include teleme-tering moorings and fixed platforms, autonomous under-water vehicles (AUVs), Lagrangian drifters, profilingfloats, and surface current measuring radar. These observa-tories complement global satellite observing networks byadding spatial and temporal resolution, or directly observ-ing the subsurface ocean, which is critical to capturingdensity stratification that often exerts significant influenceon coastal dynamics. They have gradually evolved to com-prehensive Coastal Ocean Observing Systems (COOS),
integrating a variety of data sources via networking. Datamanagement is essential and usually covered at the nationallevel. Examples are the Center for Operational Oceano-graphic Products and Services and the National DataBuoy Center in the US (under the National Atmosphericand Oceanic Administration, NOAA).
Although COOS have specific goals and features, theycannot be considered as isolated from open ocean monitor-ing efforts. Improving complementarity between coastaland open ocean observing systems can be beneficial: (a)COFS need validated boundary conditions; (b) openocean forecasting systems need to be validated in thecoastal ocean against local/regional data; (c) the benefitof coastal monitoring toward improvements in openocean prediction (‘coastal signal upscaling’) has to bequantified; (d) downstream services and user uptake inthe coastal ocean strongly depend upon the optimal func-tioning of coastal and larger-scale ocean forecastingsystems, both validated against observations.
Advances in coastal ocean observing systems
International initiatives in science policy reflect the impor-tance of coastal networks and observatories. For instance,the establishment of comprehensive COOS is beingadopted as an important component of marine strategy bythe European Commission and by most countries that areadvanced in marine science research and with economi-cally significant coastal areas (e.g. Committee on anOcean Infrastructure Strategy for US Ocean Research in2030, 2011) (European Commission 2010; European Com-mission, 2012; European Commission 2013). These newobservatories, such as the Integrated Marine ObservingSystem (IMOS, Australia), the US Integrated OceanObserving System (IOOS) and the Ocean ObservatoriesInitiative (OOI), St. Lawrence Observatory (Canada),Coastal Observing System for Northern and Arctic Seas(COSYNA, Germany), and POSEIDON System (Greece),are today discovering new insights for ocean variability.These discoveries will in turn trigger new theoretical devel-opments, increase our understanding of coastal and near-shore processes and contribute toward a more science-based and sustainable management of the coastal ocean.
New approaches include multiplatform COOS (see nextsub-section), which now allow us to characterize thecoastal ocean in quasi-real time, both in terms of theocean state and its variability at mesoscale and sub-mesoscale levels (Tintoré et al. 2013). The status ofcoastal observatories is expected to further advancethrough the integration with regional deep sea observa-tories or initiatives. Examples are JERICO (Joint EuropeanResearch Infrastructure network for Coastal Observatories,[jerico-fp7.eu/]) and ESONET (European Sea floor Obser-vatory NETwork). ESONET focuses on long-term multi-disciplinary deep sea observatories around Europe,
linking marine sensors to the shore by acoustic or cableconnection in real or near-real time at relatively high fre-quency. This approach has been tried within the frameworkof the German COSYNA project (see next section), wherecabled techniques have been implemented. Similarexchange of knowledge, tools, resources or personalsupport could enhance the durable operation of observa-tories and generate added products.
A multi-platform approach (Western Mediterraneanexample)
Studying complex coastal dynamics requires theimplementation of synergistic approaches through the com-bined use of multi-platform observing systems able toresolve a wide range of spatial and temporal scales.Recent advances have been achieved in the Western Med-iterranean Sea, where the circulation is characterized by thepresence of multiple interacting scales, including basin,sub-basin scale, mesoscale and sub-mesoscale structuresas well as coupled bio-physical processes and shelf-slopeexchanges. SOCIB, the new Balearic Islands CoastalOcean Observing and Forecasting System, is one suchsystem, a new facility of facilities, open to internationalaccess (Tintoré et al. 2013).
There are several examples of value-added by such amulti-platform, multi-scale observatory. Positive insightsconcerning the use of autonomous underwater vehicles(gliders) in synergy with altimetry, in order to monitordynamics in the Balearic Sea, have been provided (Ruizet al. 2009). Innovative strategies have been developed tocharacterize horizontal ocean flows, specifically in termsof current velocity associated with filaments, eddies orshelf-slope flow modifications close to the coast (Bouffardet al. 2010). These methodologies were applied to a seriesof glider missions carried out almost simultaneously andwell co-localized along the altimeter tracks. The valueadded by combining remote and in-situ sensors to validate,intercalibrate and improve observing data dedicated tocoastal ocean studies has been shown (Pascual et al.2010; Pascual et al. 2013). For instance, high-resolutionhydrographic fields from gliders revealed the presence ofpermanent and non-permanent signals, such as relativelyintense eddies, that were not correctly detected by standardaltimeter fields. Qualitative and quantitative comparisonswith drifters, glider, and satellite sea surface temperature(SST) observations reveal that when the new altimetryproducts are used, a better agreement is obtained (Escudieret al. 2013).
Figure 1 shows an example from the French-IndianSARAL/Altika mission with gliders along a selectedtrack in the Western Mediterranean close to Ibiza Island,where the SOCIB High Frequency (HF)-radar facility pro-vides hourly surface current velocities. Surface drifterswere also deployed in the studied region. The glider
mission (2-5 August 2013) and the passage of the satellitealong the selected track were almost simultaneous. Com-parisons (Figure 1) reveal a reasonable agreementbetween all platforms (drifter, along-track SARAL/AltiKaand HF-radar). The gradient of dynamic height measuredby the glider was only on the order of 2-3 cm, but indicatedthe presence of a coherent meander with maximum associ-ated velocities of about 20 cm/s. SARAL/AltiKa records(using 40 Hz along-track near real-time data) also capturedthe meander, with consistent size, amplitude and positioncompared to glider observations. SARAL/AltiKa was actu-ally able to capture the northern edge of the meander, whichlies on a shallow bathymetry less than 10 km from thecoast. The combination of satellite altimetry with indepen-dent in-situ data has thus demonstrated benefits for improv-ing knowledge on coastal and mesoscale dynamics.
Coastal ocean forecasting
Examples of coastal forecasting systems
An inventory of forecasting systems around the coasts of allcontinents is being kept and updated, to increase coherencebetween research developments within the frameworkof COSS-TT (see Systems Information Table, SIT,under [https://www.godae-oceanview.org/science/task-teams/coastal-ocean-and-shelf-seas-tt/]). This inventoryincludes: short description of COFS; their geographicaldomains, objectives and system status; generated products(hindcasts or forecasts and frequency of availability); dataused for assessment and quality control methods. TheSIT also describes different applications, includingCoastal Eutrophication and Hypoxia, Human Exposure toWaterborne Pathogens or Radiation, Harmful AlgalBlooms, Habitat Loss and Modification, Vulnerability toCoastal Flooding, Ocean Acidification and Food Security.Many systems share methodologies dictated by sciencedrivers and user needs. These two topics are discussed indetail in a companion paper (Kourafalou et al. 2015).Here, a few examples are showcased, chosen as representa-tive cases that highlight characteristic aspects, while offer-ing some geographical diversity. Certain systems employcoupling capabilities with atmospheric and biogeochemicalcomponents; see further discussion on coastal coupledmodels and ecosystem modeling in the companion paper(Kourafalou et al. 2015). Partial information on a fewsystem examples extracted from SIT is given in Table 1.
Example 1: A multi-nested modeling approach (China)
The Chinese Global operational Oceanography ForecastingSystem (CGOFS, [http://www.nmefc.gov.cn/cgofs_en/index.aspx]) has been recently developed by China’sNational Marine Environmental Forecasting Center(NMEFC). As a part of the CGOFS, the Yellow Sea and
the East China Sea operational oceanography forecastingsystem, CGOFS_ECS, has also been developed based onthe Regional Ocean Modeling System (ROMS, http://
myroms.org) at a horizontal resolution of ∼3-5 km and30 vertical layers (Shchepetkin & McWilliams 2005).The CGOFS_ECS model runs are separated into three
Figure 1. (Top): G-ALTIKA multi-platform experiment in the Ibiza Channel (Western Mediterranean): glider mission definition anddrifter trajectories. The vectors correspond to the surface currents derived from SOCIB HF radar (Courtesy: C. Troupin, IMEDEA,CSIC-UIB). (Bottom): Across track surface geostrophic velocity obtained during the G-ALTIKA experiment: (Left) SARAL/AltiKadata (filtered 40 Hz SLA + SMDT-MED-2014) and (Right) glider data (Dynamic height computed with a reference level of 600 m).
parts: (a) 10 years climatology run for spin-up; (b) 2000-2012 hindcast run; (c) forecast run (2013–present).
The climatology run is forced by monthly mean clima-tology wind stresses, net fresh water fluxes, surface heatfluxes from COADS (Diaz et al. 2002). Fields for openboundary conditions are from the monthly mean climatol-ogy Simple Ocean Data Assimilation (SODA) datasets(Carton & Giese 2008). Initialized from this run, the hind-cast simulation is forced by the 6-hourly forecasted productsfrom NOAA’s National Centers for Environmental Predic-tion (NCEP) Climate Forecast System Reanalysis (CFSR)(Saha et al. 2010). Open boundary conditions are derivedfrom the monthly mean of each year of the SODA datasetsand are complemented by harmonic constants of 10 tidalconstituents extracted from the Oregon State UniversityTidal Data Inversion [http://volkov.oce.orst.edu/tides/](Egbert & Erofeeva 2002). Monthly mean climatology dis-charges of the Yangtze River are included. Buoy floats in themodel domain are used for hindcast evaluation.
For the forecast run, the model started from the hindcastrun on January 1st, 2013, and was forced by the NMFEC
atmosphere forecasting system based on the WRFmodel (Weather Research and Forecasting) (Skamarocket al. 2005). CGOFS_ECS runs daily for 6 days (1-daynowcast and 5-days forecast). Daily updated 120-hourforecasting products (see examples on Figure 2) are usedfor: open boundary conditions for high resolution coastalocean models (forecasting oil spill or red/green tide,marine search and rescue); navigation, fisheries manage-ment, marine environmental protection.
Example 2: Data assimilative coastal modeling (Japan)
The Meteorological Research Institute (MRI) of JapanMeteorological Agency (JMA) has been developing theMOVE/MRI.COM-Seto coastal monitoring and forecast-ing system, consisting of a fine-resolution (2 km) coastalmodel and an eddy-resolving (10 km) data-assimilativemodel. The coastal model (50 vertical layers) is based onthe MRI Community Ocean Model (MRI.COM) and isone-way nested into the Western North Pacific (WNP)model (Tsujino et al. 2011). Four-dimensional variational
Table 1. Examples of systems (alphabetic order per Region) featured in the Systems Information Table maintained by the Coastal andShelf Seas Task Team of GODAE/Oceanview. (Region I: Americas; Region II: Asia and Australia; Region III: Europe).
Acronym System Name Country Domain(s)
I
ESTOFS Extratropical Surge and Tide OperationalForecast System
USA US East, Gulf of Mexico andWest Coasts, up tothe Gulf of Alaska.
FKeyS-HYCOM
Florida Straits, South Florida and Keys HybridCoordinate Ocean Model
USA Florida Straits and the South Florida coastal andshelf areas
NWPS Nearshore Wave Prediction System USA Coastal Waters of all US territoriesNYHOPS New York Harbor Observing and Prediction
SystemUSA Coastal waters of the Middle Atlantic Bight on
the East Coast of the US (<200 m deep).OFS Operational Forecast System USA All major estuaries and coastal systems of the
USP-Surge Probabilistic Surge USA Coastal and overland areas of all US territoriesREMO Oceanographic modelling and observation
networkBrazil Western Equatorial and South Atlantic Ocean
SLGO St. Lawrence Global Observatory Canada Gulf of St. LawrenceWCNRT West Coast Near Real Time Data Assimilation
SystemUSA West US Coast, California Current System
II
CGOFS Chinese Global operational OceanographyForecasting System
China Global and Regional seas around China
ESROM_MOM Regional ocean modelling system South Korea East Sea (Japan Sea)eReefs eReefs Marine Modelling Australia Australian coastal marginsMOVE/MRI MRI Multivariate Ocean Variational Estimation
System / MRI Community Ocean ModelJapan Global, North Pacific, Western North Pacific,
Coastal region around JapanYS_ROMS Korea operational oceanography system South Korea Yellow Sea and East China Sea
III
AFS Adriatic Forecasting System Italy Adriatic Sea and Northern Ionian SeaCOSYNA Coastal Observation System for Northern and
Arctic SeasGermany North Sea, German Bight, German Wadden Sea
MFS Mediterranean Ocean Forecasting System Italy Mediterranean SeaNEMO-FOAM NEMO FOAM Operational Modelling United
KingdomEuropean Northwest continental shelf
POSEIDON Regional monitoring and forecasting system Greece Aegean and Mediterranean SeasPREVIMER PREVIMer Coastal observations and forecasts France Bay of Biscay / English Channel / Northwestern
(4DVAR) data assimilation is applied to WNP, based on a4DVAR version of the MRI Multivariate Ocean VariationalEstimation system (MOVE-4DVAR). The coastal model is
initialized using the 4DVAR analysis fields of the WNPthrough incremental analysis updates after interpolationto the finer grid (Bloom et al. 1996).The one-way nesting
Figure 2. 24-hour forecast products of the CGOFS_ECS system on December 22, 2013. (a) Currents at the surface; (b) Currents at the 10m layer; (c) Temperature at the 20 m layer; (d) Temperature at the 50 m layer.
s132 V.H. Kourafalou et al.
Dow
nloa
ded
by [
Uni
dad
de R
ecur
sos
de I
nfor
mac
ion
Cie
ntif
ica]
at 0
0:47
18
May
201
6
technique is adopted in MOVE/MRI.COM-Seto. Thecoastal model is driven by 3-hourly atmospheric conditions(from the JMA atmospheric operational Meso-ScaleModel, MSM) and 6-hourly radiative heat fluxes (fromthe Global Spectral Model, GSM).
MOVE/MRI.COM-Seto will be operated at JMA andwill be mainly used for its warning system around thecoastal region and possibly for fisheries, ship navigation,and marine leisure. There is also collaboration with JapanAerospace Exploration Agency (JAXA), planning a newaltimeter mission (Coastal and Ocean measurementMission with Precise and Innovative Radar Altimeter,COMPIRA). COMPIRA will carry a wide-swath altimeter(Synthetic Aperture Radar Height Imaging Oceanic Sensorwith Advanced Interferometry, SHIOSAI). To develop highaccuracy COMPIRA coastal and open ocean products,JAXA has launched the ‘COMPIRA coastal forecast coreteam’. In this framework, Observing System SimulationExperiments (OSSEs; see also next section) will be per-formed to evaluate the effectiveness of the new satellitedata and the feasibility of developing high accuracy pro-ducts using data assimilation schemes for coastal regions.
Example 3: Resolving intra-tidal cycles (Germany)
TheCoastal Observing SYstem forNorthern andArctic Seas[COSYNA; http://www.hzg.de/institute/coastal_research/cosyna/] has been deployed in the German Bight, integratingnear real-timemeasurementswith numericalmodels andpro-viding continuous coastal ocean state estimates and fore-casts. COSYNA, which is operated by the Institute ofCoastal Research, Helmholtz Zentrum Geesthacht (HZG),showsmany similarities with advanced coastal observatoriesin the US and Europe (e.g. Glenn & Schofield 2009;Proctor & Howarth 2008). It consists of observationalnodes, a data management system and data assimilationcapabilities, streamlined towards meeting the needs forhigh quality operational products in the German Bight. Theindividual in-situ observing subsystems used are: FerryBox,gliders, buoys and HF-radar. The forecasting suite includesnested 3D hydrodynamic models running in a data assimila-tion mode, forced with meteorological forecast data.
Unlike most systems assimilating HF-radar data, whichare concerned with low-pass filtered surface velocitymeasurements, COSYNA focuses on intra-tidal scales,which can be justified by the need to a) develop a betterknowledge on the short-term coastal ocean variability,and b) enhance quality of data needed for special coastaloperations. The blending of data and models (see alsonext sub-section) uses a spatio-temporal optimal interp-olation (STOI) which enables dynamically consistentsmoother within an analysis window of one or two tidalcycles. This method maximizes the use of available obser-vations, as a step towards ‘best surface current estimate’.Patchy observations over part of the German Bight
sampled every 20 mins from three WERA radars are usedto prepare 6-hr and 12-hr forecasts. COSYNA modelingproducts also include regular maps of wind, waves, salinity,and temperature. The latter two are enhanced by the assim-ilation of FerryBox data (Stanev et al. 2011).
Example 4: National initiatives in coastal oceanforecasting (NOAA/USA)
Coastal wave and surge modeling systems typically makeuse of phase-averaged spectral wave models such asSWAN (Simulating Waves Nearshore), and increasinglyWAVEWATCH III, coupled to varying degrees with circula-tion models such as SLOSH, ADCIRC, FVCOM (FiniteVolume Coastal Ocean Model) and SELFE (SemiimplicitEulerian-Lagrangian Finite-Element), typically run in two-dimensional, depth-integrated mode (Booij et al. 1999;Tolman et al. 2002; Jelesnianski et al. 1992; Luettich et al.1992; Chen et al. 2003; Zhang & Baptista 2008). The USfederal agency that oversees operational oceanic prediction(NOAA) has developed operational guidance systemswhere these models are currently run in uncoupled (orone-way coupled) mode: ADCIRC-based Extra-tropicalSurge and Tide Operational Forecast System (ESTOFS),the SLOSH-based Probabilistic Hurricane Storm Surge(P-Surge) and the SWAN/WAVEWATCH III-basedNearshore Wave Prediction System (NWPS) (Feyen et al.2013; Taylor & Glahn 2008; Van der Westhuysen et al.2013). In these systems, aspects of physical phenomena areshared between models (e.g. ESTOFS and P-Surge waterlevels are included in the NWPS wave model, see Figure 3),but there is no process feedback. An example of a fully-coupled wave-surge system is the ADICRC and SWAN-based ADCIRC Surge Guidance System (ASGS). Here thesurge model transfers water levels and depth-integrated cur-rents to the wave model, which, in turn, transfers wave radi-ation stresses and enhanced bed friction to the surge model.
Coastal three-dimensional baroclinic circulation model-ing systems are designed to provide guidance on waterlevels, currents, salinity and temperature. Examples ofsuch systems are NOAA’s national network of OperationalNowcast and Forecast Hydrodynamic Model Systems(called OFS). An OFS consists of the automated integrationof observing system data streams, hydrodynamic modelpredictions, product dissemination and continuousquality-control monitoring. Within these systems, hydro-dynamic models such as ROMS, FVCOM and SELFEare driven by real-time data and meteorological,oceanographic, and/or river flow rate forecasts, receivingboundary conditions at the coastal shelf from NOAA’sGlobal-RTOFS (Real-Time Ocean Forecast System),based on HYCOM [HYbrid Coordinate Ocean Model,http://hycom.org] (Chassignet et al. 2007).
To promote the next generation of operational forecast-ing, NOAA’s Integrated Ocean Observing System [IOOS;
http://www.ioos.noaa.gov/] has established elevenRegional Associations that form a national network ofRegional Coastal Ocean Observing Systems (RCOOS).These state-of-the-art observational and complimentarymodeling activities are supported by consortia of federal,state, academic and commercial partners.
Example 5: A coupled coastal system (USA)
This example is based on one of the RCOOS mentionedabove, namely the Southeast Coastal Ocean Observing
Regional Association (SECOORA). An integrated high-resolution, three-dimensional, coupled (ocean-atmos-phere-wave) Nowcast/Forecast system has been developedfor the Northwest Atlantic Ocean by North Carolina StateUniversity (NCSU). Covering the entire US east coastalocean, the Gulf of Mexico and Caribbean Sea, the systemis implemented based on the Coupled Ocean-Atmos-phere-Wave-Sediment Transport (COAWST) modelingsystem (Warner et al. 2010). COAWST couples ROMS,WRF and SWAN models representing the ocean, atmos-phere, and wave environments. ROMS/SWAN is spatially
Figure 3. Implementations of the Nearshore Wave Prediction System over the US Gulf Coast (rectangles, top), and results over thedomain of the New Orleans Weather Forecast Office, including probabilistic surge levels from NOAA’s P-Surge.
collocated with the WRF domain (7-10 km grid, fineenough to resolve atmospheric forcing from tropicalcyclones) (Halliwell et al. 2011). Boundary and initial con-ditions are provided by the global HYCOM.
These three models were coupled using the ModelCoupling Toolkit (MCT), resulting in a COFS that exhibitsseveral advantages over global forecasting systems (Larsonet al. 2004; Jacob et al. 2005). These include: (a) finer res-olutions in both horizontal and vertical directions that canbetter resolve regional and coastal processes (Hurlburt &Hogan 2000); (b) fully coupled model physics thatinclude the interactions/feedbacks among ocean circula-tion, marine meteorology, and ocean waves; (c) animproved representation of coastal/shelf dynamics (e.g.tides). The coupled system performs routine nowcast and3-day forecast on daily basis [http://omgsrv1.meas.ncsu.edu:8080/ocean-circulation-useast2] with an example pro-vided in Figure 4. Near-real time model predictions arevalidated against HF radar surface currents, NOAA sealevel data and buoy measurements. Interactive functionsinclude: visualizations of user defined virtual station
profiles or hydrographic transects and 72-hour surface tra-jectory ‘virtual particle’ simulations.
Example 6: From the ocean to the reef scale (Australia)
Reaching reliable model forecasts in fine coastal scalesrequires careful downscaling (see also Kourafalou et al.2015). The eReefs project [http://www.emg.cmar.csiro.au/www/en/emg/projects/eReefs.html] is highlighted hereas an example associated with a unique set of challenges(Schiller et al. 2014). This initiative aims to provide aninformation system, underpinned by models, for theiconic Great Barrier Reef (GBR) on Australia’s north-east coast. The GBR is the longest stretch of coral reefin the world, one of the seven natural wonders of theworld, a UNESCO world heritage site and home to abun-dant biodiversity. Reef cover has continued to decline overthe last several decades, due to the effects of cumulativestresses, primarily nutrient loads from terrestrial runoff,Crown-of-Thorns Sea-star infestations and damage fromtropical cyclones (Brodie & Waterhouse 2012). Some of
Figure 4. Concurrent snapshots of the coupled COAWST model simulated: (upper panels) sea level air pressure, surface wave height anddirections, surface ocean velocity and sea level; (bottom panels) 10 m wind, sea surface temperature and sea surface salinity.
these stresses can be mitigated by targeted managementstrategies (e.g. nutrient load input), whereas otherscannot (e.g. extreme weather events). The eReefssystem, therefore, aims to provide managers relevantinformation to assist in the development of informed miti-gation strategies to improve reef health. Consequently, anymodels contributing to the overall modeling system mustspan scales (from the catchment, through estuaries,across the lagoon, over the reef matrix and across theshelf to the deep ocean) and disciplines (catchment mod-eling, hydrodynamics, waves, sediment transport andbiogeochemistry).
Nesting from the global to the reef scale requiresseveral downscaling nests: eReefs (Figure 5) employs a 4km ‘bridging model’ in global products and a 1 km regionalmodel, with nested re-locatable models of estuaries/reefs(100s of meters) (Herzfeld 2009; Herzfeld et al. 2011;Herzfeld & Andrewartha 2012). The reef matrix can gener-ate fine scale structure in the flow, which can feed back tothe larger scale (Wolanski & Hamner 1988; Wolanski et al.1996; Wolanski et al. 2003a; Wolanski et al. 2003b). Tooptimize runtime, complex curvilinear grids utilizing
branching are employed to represent only areas of interest.Additionally, an unstructured coordinate system is utilizedto ‘house’ state variable matrices within the model; thisfacilitates the representation of wet cells only in the statevector, which improves computational efficiency. Althoughthe coordinate system is unstructured, the model is basedon finite differences. The reef creates large topographic gra-dients, and individual reef lagoons are isolated from thesurrounding waters by exposed fringing reefs at everytidal cycle (tidal range can be 6 m at the coast). Differentialheating/evaporation can significantly modify water proper-ties in these isolated lagoons, which feed back to the largerscale when the reef becomes wet again at high tide. Thesedynamics promote the use of a ‘z’ vertical coordinatesystem with true wetting and drying.
Model and data integration
The integration of the multiplatform observing and fore-casting systems described previously is necessary toachieve a comprehensive description of the dynamics in
Figure 5. A nested approach to traverse scales (Greater Barrier Reef, color is bathymetry). The regional configurations span a portion ofthe Coral Sea between Papua New Guinea and Queensland, Australia. The local models cover the Fitzroy River estuary and Moreton Bayoff Brisbane. This multi-nesting downscaling cascade achieves resolution at the local scale, while maintaining open boundary nesting ratiosof less than 5:1 (2 regional nests at 4 km and 1 km, allowing local nests to go below boundary resolutions of ∼200 m). Outer models areOceanMAPS (Oke et al. 2008) at 10 km resolution for the ocean initial and boundary conditions, and ACCESS for surface fluxes [http://www.bom.gov.au/nwp/doc/access/NWPData.shtml].
the coastal ocean, where several phenomena are controlledby small-scale changes in time and space. Dynamicalinterpolation of data on the regional/coastal scale canplay a key role in synthesizing the data derived infor-mation, toward monitoring and predicting variability fromdays/weeks to seasonal/decadal.
The past decade has witnessed the establishment ofnumerous regional coastal ocean observatories around theworld. These have prototyped different approaches to coor-dinate multiple observing technologies for real-time coastalocean monitoring, and support coastal forecasting.Examples include the US/NOAA Regional Associationsunder IOOS and the European Copernicus MyOceanproject [http://MyOcean.eu], closely linked to productsfor decision-making to improve safety, enhance theeconomy, and protect the environment.
Initiatives to provide specialized data sets for COFSinclude the Group for High Resolution SST (GHRSTT).An example is given for the Great Barrier Reef (eReefs,described above) based on a 1-year simulation (Figure 6).There is a tendency for the model to underestimate theSST by between 0.5 and 1.0°C during the wet season(November - April). However, at the onset of the dryseason, the bias reduces to <0.5°C and most of the valueslie within the standard error of the GHRSST observations.Other efforts that are of particular utility for coastal oceanforecasting include Coastal Altimetry that is developingmethods to expand the capabilities of standard altimetryproducts in the coastal areas (Cipollini et al. 2012).
A particular challenge is to combine observations andmodels in areas of strong coastal to offshore interactions.A range of appropriate scales needs to be satisfied, with
Figure 6. High resolution model to data monthly comparison for the Great Barrier Reef. Two-dimensional histogram of binned GHRSSTdata vs. model SST output during the period September 2010 - August 2011. Color denotes the number of observations (frequency) of aparticular data/model combination. Dark red denotes the highest mass. The black line denotes the 1:1 relationship between the GHRSSTdata and model output. This analysis gives more insight than a typical scatter plot.
models and data having complementary resolution in timeand space. An example is given in Figure 7 for the impactof Florida Current meandering (a component of the GulfStream western boundary current system) on coastalflows along the South Florida coastal areas (which hostthe Florida Keys National Marine Sanctuary, the largestreef system in the continental US). As the Florida Currentmeanders between Florida and Cuba, cyclonic frontaleddies undergo synergistic changes that influence coastalflows, with implications on the reef ecosystem (Kourafalou& Kang 2012; Sponaugle et al. 2005). Two such eddies areevident in the fields predicted by the Florida Straits, South
Florida and Florida Keys (FKEYS) HYCOMmodel, whichis nested in the data-assimilative Gulf of Mexico (GoM)regional HYCOM model at a resolution of ∼900 m,which is four times higher than the regional model andeight times higher than the global HYCOM model.Together with other attributes that enhance coastal perform-ance, the location and eddy size in the FKEYS fields are invery good agreement with the observed cyclone in theupper Florida Keys, where a HF-radar (WERA) is main-tained (data provided from [http://iwave.rsmas.miami.edu/wera/] and specific eddy event discussed in Parks et al.2009). This is not the case for the regional model, which
Figure 7. (Top left): Sea Surface Height and near-surface current around South Florida and the Florida Straits from the FKEYS nestedmodel (21 January 2005); the meandering of the Florida Current is depicted with two mesoscale cyclonic eddies (north of Cuba and east ofthe upper Florida Keys). (Top right): Observed surface currents from the WERA HF-radar (same date) showing the cyclonic eddy off theupper Florida Keys. Detail within the WERA covered area of the nested FKEYS model (bottom left) and the outer Gulf of Mexico regionalmodel (bottom right); the regional model does not represent the observed eddy. Data from [http://iwave.rsmas.miami.edu/wera/]
does not resolve this eddy and, therefore, misses the coastalsouthward flow along the Florida Keys reef system.
Land to sea interactions are another topic of particularimportance in coastal forecasting, with challenges associ-ated with both the monitoring of riverine discharges, aswell as the correct representation of river plume dynamicsand the resolution of the related buoyancy-driven flows(Schiller & Kourafalou 2010). When boundary currentsinterfere with the evolution of river plumes, additionalcomplexities arise, presenting both monitoring and model-ing challenges. An example is the interaction between theMississippi River plume and the Loop Current (GulfStream system branch in the Gulf of Mexico), whichdepends on both coastal circulation and the complex,large scale boundary current and eddy field (Schiller et al.2011). This process is not well represented in regionaland global models, which are of coarser resolution and gen-erally rely on relaxation to climatology for the salinity field,thus unable to replicate river plume observations as well ascoastal models (Kourafalou & Androulidakis 2013).
Increasingly, model validation is being done within thecontext of standardized test beds such as NOAA’s JointHurricane Testbed (JHT) and the Coastal and Ocean Mod-eling Testbed (COMT) (Rappaport et al. 2012; Luettich Jret al. 2013). These make use of standardized metrics, testcases and advanced IT infrastructure for sharing and com-paring model results. An emerging industry standard for
coastal model validation is the Interactive Model Evalu-ation and Diagnostics System (IMEDS), developed withthe support of the IOOS COMT (Devaliere & Hanson2009). Evaluations of wind, wave and water level datacan be made on large temporal and spatial scales to statisti-cally reduce large volumes of model estimates to meaning-ful measures of prediction skill. A variety of time-serieserror metrics (such as root-mean-square error, bias, scatterindex) are included. The system features three statisticalapproaches, namely Temporal Correlations (TC), Quan-tile-Quantile (QQ), and Peak Event (PE) analyses, repre-senting industry benchmarks for operational modelvalidation.
In addition to using observations for model evaluation,their direct insertion in models is achieved through dataassimilation methods, which have to be specificallyadapted for coastal systems (Kourafalou et al. 2015).Figure 8 presents an example of the impact of assimilatingin-situ data from AUVs and ships of opportunity in theROMS model with 4D variational data assimilation(4DVAR, Moore et al. 2011a) that is part of MARACOOS,the IOOS Regional Association for the Mid-Atlantic Bight(MAB). Figure 8(a) shows model-estimated temperature atthe seafloor when only satellite SST and coastal-correctedalong-track altimeter sea surface height (SSH) are assimi-lated. This pattern overestimates the extent of the MAB‘cold pool’ (temperatures less than 11°C). Figure 8(b) is
Figure 8. Bottom temperature on the Mid-Atlantic Bight shelf (in water depths < 1000 m) estimated by Rutgers University’s near-real-time ROMS model with 4DVAR assimilation for 22 Sep 2013. (a) Analysis with assimilation of satellite data only. (b) Analysis with sat-ellite data plus HF-radar surface currents and in-situ temperature and salinity from underwater profiling gliders. Red circles indicate bottomtemperature observed by 4 gliders in the 15 days preceding the analysis time.
Journal of Operational Oceanography s139
Dow
nloa
ded
by [
Uni
dad
de R
ecur
sos
de I
nfor
mac
ion
Cie
ntif
ica]
at 0
0:47
18
May
201
6
the corresponding analysis when MARACOOS gliderbottom temperature data are added to the assimilation. Asevident in Figure 8(b), in-situ data assimilation decreasesthe eastern extent of the cold pool and shows a filamentof warm water at the shelf break that subsequentlyadvects southwestward, beneath the shelf/slope front, incloser agreement to observations. The assimilated sub-surface data were essential to overcome the limitation ofremote sensing imagers, which miss the cold pool watersthis time of the year (early fall), when surface temperaturesare warm everywhere.
Array design methods
The optimization of observational networks is an impor-tant, cost effective, aspect of integrating data and models.Observation array design refers to numerical methodsused to perform analyses of the performance of observationnetworks, with the purpose of evaluating existing arrays,testing alternate configurations, or estimating the impactof future deployments or instruments.
The most common approach for array design involvesthe use of data assimilation into a numerical model, inthe framework known as Observing System Experiments(OSE), or Observing Systems Simulation Experiments(OSSE) (Oke et al. 2015). In OSEs, a set of data assimila-tion experiments is performed, using actual observations.The best simulation is the one into which all availableobservations are assimilated, achieving the largest errorreduction with respect to a simulation in which no obser-vation is assimilated. The performance of a specific obser-vational array is then estimated by running an experimentin which all the observations are assimilated, except forthe ones from the array under study. The change in errorreduction between that latter experiment and the exper-iment in which all observations are assimilated allowsquantifying the impact of that specific network. Althoughmany studies based on OSEs have been performed to testthe impact of various components of the global oceanobserving system (see Oke et al. 2015), such work hasstarted more recently for testing the impact of networksdedicated to the coastal ocean. For instance, the positiveimpact of ∼3 months of glider observations, even after afew months, on the forecast of the Eastern MediterraneanSea was demonstrated by Dobricic et al. (2010). Morerecently, an OSE took place in real time to show thebenefit of adaptive sampling by sea gliders for constraininga model of the Ligurian Sea (Mourre & Alvarez 2012).
Other studies based on the assimilation of actual obser-vations also focus on the impact of various observation net-works or data type, without performing actual OSEs. Thisis especially the case of studies using 4DVAR data assim-ilation, which allows identifying the impact of a set ofobservations using the adjoint and tangent linear of theocean model. A representer-based method was used to
assess the improvement brought to a forecast model offOregon by satellite observations of SST and altimetry,and by HF-radar observations (Bennett 2002; Kurapovet al. 2011; Yu et al. 2012). The sensitivity of simulatedcoastal current to different observation types or arraysetting was tested with a model of the California CurrentSystem, where a single 4DVAR simulation could be usedto test the impact of various observation networks in afashion similar to OSEs (Moore et al. 2011b).
OSSEs are based on the same principle as OSEs, exceptthat the observations used during data assimilation are nottrue observations, but are sampled from a second, realisticsimulation. This method avoids bias (which can compro-mise the results of OSEs) and allows testing instrumentsthat do not exist yet, or testing alternate deployment strat-egies. Several studies using the OSSE framework havebeen performed to test the impact of various observing net-works or deployment strategies in regional/coastal oceans,especially in the Mediterranean Sea: sampling strategies fortemperature profiles and for a regional observing systemcombining moorings and gliders have been tested(Raicich & Rampazzo 2003; Alvarez & Mourre 2012).An approach comparable to OSSEs, but also taking advan-tage from a representer-based approach using a variationaldata assimilation scheme, was used to test various scenariosof glider deployment in the New York Bight (Zhang et al.2010). Theoretically, the OSSE approach requires carefultesting of the components of the system (Nature Runfrom which pseudo-observations are extracted, assimilativemodel, data assimilation procedure), to make sure that itproduces realistic impact assessment. In particular, itrequires that the Nature Run and the assimilative modelare substantially different (type of model and attributes),which has not been the case in most of ocean OSSEs sofar. Such a carefully evaluated OSSE system prototypehas been recently demonstrated over the Gulf of Mexico(Halliwell et al. 2014).
The OSE/OSSE methodologies are very efficient inproviding a quantitative assessment of the impact of aspecific network. However, since they imply the use ofseveral data assimilation experiments, their implementationhas a high computational cost. They also face the same dif-ficulties as any data assimilation experiments performed incoastal/regional areas, due to the superposition of variousspace and time scales. In particular, the estimation of themodel error covariance matrix is not trivial. Coastalocean processes are strongly constrained by topography,so that error structures, unlike in the open ocean, cannotbe considered anisotropic. Ensemble approaches, such asthe Ensemble Kalman Filter (EnKF), is a commonapproach for representing non-linear error evolution thatis well adapted to the coastal ocean (Evensen 1994). Thesystem performance can then be estimated by the reductionof the ensemble spread due to the assimilation of obser-vations. Such approach has been used in an OSSE
s140 V.H. Kourafalou et al.
Dow
nloa
ded
by [
Uni
dad
de R
ecur
sos
de I
nfor
mac
ion
Cie
ntif
ica]
at 0
0:47
18
May
201
6
framework to test the impact of various altimetric obser-vations scenarios, including the use of wide-swath altimetersuch as the future Surface Water Ocean Topography(SWOT), over the North Sea (Mourre et al. 2006; LeHénaff et al. 2008).
Alternative methods exist in observation array designthat do not involve data assimilation. Some of thesemethods test the ability of an observation array to capturethe variability of the ocean signal. Deploying in-situ obser-vation arrays at locations associated with high amplitude inthe spatial EOF of the dominant modes was found to be anefficient approach for reconstructing the full 3D signal overthe Massachusetts Bay (Yildirim et al. 2009). Thisapproach allows, for example, testing the impact of thenumber of moorings on the performance of the reconstruc-tion. Other approaches are by design closer to a data assim-ilation approach, where the observations are used toconstrain the error made otherwise by the model. TheRepresenter Matrix Spectrum approach (RMspec) aims atquantifying the number of model error modes (estimatedwith an ensemble of simulations and in the observationspace) a specific network can detect, taking into accountthe observation error covariances (Le Hénaff et al. 2009;Oke et al. 2015). A comparable approach, based on therepresenter method from the variational approach, andwithout data assimilation, was used to illustrate the positiveimpact of surface velocity measurements off Oregon forconstraining the coastal upwelling system (Kurapov et al.2009). Similar concepts from estimation theory were alsoinvoked, in the Kalman formalism, for assessing a coastalnetwork of HF-radar, tide gauges and altimetry in theGerman Bight area, showing the importance of continuoustide gauge measurements (Schulz-Stellenfleth & Stanev2010). Other alternative approaches include using theBest Linear Unbiased Estimator (BLUE), adapted frommeteorology and based on the exploration of uncertaintiesof both the ocean state and the observations; it wasimplemented to test an array of moorings in the ColumbiaRiver estuary (Frolov et al. 2008). Although the develop-ment and implementation of observation array assessmentand design in the regional/coastal ocean are fairly recent,this research field is very active now.
Based on the RMspec method, alternative strategieshave been evaluated for collecting vertical temperature pro-files on fishing nets using the French RECOPESCAnetwork [http://sih.ifremer.fr/], Leblond et al. (2010).
For the RMspec analysis, a 50-member model ensem-ble has been carried out using the MARS3D ocean model[http://wwz.ifremer.fr/mars3d] for 2006 (Lazure &Dumas 2008). The ensemble has been generated by per-turbing atmospheric forcings, bottom friction, turbulent-closure coefficient, and the light extinction coefficient(mainly in river plumes). Several scenarios of in-situ obser-vations of opportunity (depending on fishing activity) havebeen compared with each other. For each network, the
number of representer matrix eigenvalues higher than 1represents the number of model error modes which thenetwork can detect. The RMspec analysis (not shown)revealed the importance of a geographically balanced dis-tribution of measurements, including regions such as thewestern Channel and the South of the Bay of Biscay(Northeast Atlantic), as opposed to denser offshoremeasurements, which are associated with lesser uncertain-ties (and lesser variability).
Probabilistic approaches and risk assessment in thecoastal ocean
As in the open ocean, the most straightforward approach tosupport coastal ocean forecasting and applications is basedon two steps: deterministic, realistic numerical modelingand validation with respect to observations; and optionaldata assimilation, which in turn enables forecasting. Anexample of such an approach to risk assessment is theMOVE/MRI.COM-Seto system (described above), target-ing coastal surge issues. A case study was conducted foran unusual tide event that occurred in September 2011and caused flooding at several coastal areas south ofJapan. Figures 9b-d shows time series of sea levelanomalies at three tide-gauge stations along the southcoast of Japan. Significant sea-level rise in the end of Sep-tember corresponds to the unusual tide event. The assimi-lated results succeed in reproducing the observed sea-level rise. The model results reveal that coastal trappedwaves induced by a short-term fluctuation of the KuroshioCurrent around 34N, 140E caused the significant sea-levelrise at south coast of Japan (Figure 9(a)). Forecast exper-iments starting from assimilated initial conditions (notshown) have indicated that this event is predictable half amonth ahead.
Besides the above classical approach, probabilisticapproaches could provide an interesting alternative in thecoastal ocean. Probabilistic forecasting is used to accountfor uncertainty in a dynamical system by generating arepresentative sample of the possible future states. Due tothe chaotic nature of the dynamics in the combined atmos-pheric, coastal and wave system, small errors in initialstates or model parameterizations can grow to significantforecast inaccuracies in time. To address this, multipleruns with either realistic perturbations of the initial (analy-sis) state, or with different models (or model formulations)are made. The resulting ensemble of results of a given par-ameter can be analyzed either in terms of their mean (typi-cally of greater skill than any one member), or their spread(indicative of the forecast uncertainty). Given a sufficientlylarge ensemble, the relative frequency of occurrences of anevent from the ensemble can be used to estimate the prob-ability of that event.
In many coastal regions of the world, there may not beenough observations to reliably estimate uncertainties of
numerical models, to assimilate in those models, and toenable deterministic forecasts (Schiller et al. 2015). There-fore, it can be expected that probabilistic approaches will bewidely used in the future, complementing the ‘determinis-tic’ approach, for quantifying uncertainties in coastal pro-ducts, and for providing probabilistic forecasts. Inaddition, for authorities and service companies involvedin applications, such as coastal flooding, fish stock manage-ment or surface drift predictions, probabilistic productshave the potential to facilitate crisis-time decision-making, and the longer term policies aimed at the mitiga-tion of risks, with respect to using deterministic, best-esti-mate products alone.
An example of probabilistic forecast system applied tostorm surge is NOAA’s P-Surge. Due to the initial state andmodeling uncertainties in tropical cyclones, NOAA’sNational Hurricane Center (NHC) has been utilizing prob-abilistic storm surge and coastal inundation forecast gui-dance for the past decade. With P-Surge, thousands ofSLOSH model runs are made, forced by hurricane model
input parameters from normal distributions centered onthe current NHC official forecast, but with standard devi-ations based on historical errors in official NHC track andintensity forecasts. These include along-track (forwardspeed) and cross-track errors, variation in the radius ofmaximum wind and variation in intensity.
Besides the above examples, the use of probabilisticmethods is quite embryonic in the coastal forecasting com-munity, even while they can be viewed as an extension ofthe now familiar Ensemble-based approaches (Chen et al.2009). Other techniques include Monte-Carlo Markovchain (MCMC), Bayesian inference approaches, fuzzylogic, Bayesian hierarchical networks and methods to vali-date probabilistic forecasts, e.g. Brier score (Robert &Casella 2004; Pelikan et al. 2005; Jolliffe & Stephenson2011). Several current uses of probabilistic methods arevery relevant: coastal flooding and sea-level surges,forecasting extreme events and surface drift forecasting(Apel et al. 2006; Purvis et al. 2008; Abramson et al.1996; Rixen et al. 2008; Vandenbulcke et al 2009).
Figure 9. (a) Sea surface height (contour) and anomaly (color) on 28 September 2011 obtained from MOVE-4DVAR (unit in cm). Timeseries of sea-level anomalies at three tide-gauge stations: (b) Sumoto (SM), (c) Uwajima (UW), and (d) Aburatsu (AB). Gray thick linesdenote observed sea-level anomalies and black lines with open circles are results of the coastal model. Sea-level rise in the end of Septembercorresponds to the unusual sea level event mentioned in text. Sea-level anomalies for tide-gauge data, which are defined as deviations fromastronomical tide including seasonal sea level change, are corrected for barometric pressure using sea level pressure obtained from the Japa-nese 25-year Reanalysis data (Onogi et al. 2007). Those for MOVE-4DVAR are anomalies from daily mean climatological sea level fromlong-term reanalysis data (Usui et al. 2006)
s142 V.H. Kourafalou et al.
Dow
nloa
ded
by [
Uni
dad
de R
ecur
sos
de I
nfor
mac
ion
Cie
ntif
ica]
at 0
0:47
18
May
201
6
Conclusions
Coastal Ocean Forecasting Systems (COFS) are operatingin many regions of the world’s coastal ocean, providingestimates of diverse marine variables of interest andserving local needs. At the same time, they obey similarprinciples and face similar challenges in data andmethods. A challenge in itself is the fact that the coastalresearch community is traditionally more fragmented thanthe global ocean community.
Modern research and monitoring activities in ocean-ography have resulted in a rapidly increasing number ofobserving systems. Their networking, by establishingappropriate infrastructures, is capable of providing continu-ous and sustainable delivery of high quality environmentaldata and information products related to the coastal marineenvironment. However, the present-day situation is thatcoastal observations are usually carried out by individualcountries, in isolation, and sometimes in a non-sustainableway. Dissemination strategies also vary considerablybetween different countries. End-to-end coastal monitoringfrom data acquisition to data dissemination is oftenmissing. Similarly, available products for forcing coastalmodels have great limitations. Observations might notresolve the desired scales, while outputs of larger scalemodels (used for initialization and boundary forcing, bothlateral and from the atmosphere) can be inadequate.
The need to increase the coherence and the sustainabilityof dispersed national coastal observatories is being recog-nized by putting in place common framework and standards.IOOS in the US and JERICO in Europe give some goodexamples, extending global and regional initiatives to thecoastal ocean. Such efforts are important to streamline datagathering from coastal observatories to products suitablefor decision- and policy-making in the socioeconomicallyvital and often environmentally stressed coastal regions.
New and strategic technologies need to be identifiedand implemented in the next generation coastal observa-tories. For example, using new satellites (e.g. SARAL/AltiKA, SWOT) or land-based networks (e.g. HF-radars)helps achieve better sampling in time and space. Usingautomated platforms and sensors systems, as well as ensur-ing autonomy over long time periods, are also desirable.
In addition to adequate observations, the integration ofmulti-platform observatories with models that resolvecoastal dynamics is clearly a key feature of successfulCOFS. Such integrated systems must be linked to largerscale systems toward the achievement of seamless datasets, nowcasts and forecasts from the global to the littoralscale. The ultimate goal is for COFS to demonstrateadded value on open ocean systems, in the context of varia-bility over short and long scales.
Emerging capabilities have been discussed. In particu-lar, OSEs and OSSEs, adapted for the coastal ocean,provide a rigorous, cost-effective approach for the
optimization of existing coastal observing systems and theplanning of future ones (e.g. SWOT satellite mission).Coastal/regional OSEs and OSSEs will be most helpful to:(a) establish how well the different platforms contribute tocharacterize coastal ocean state and variability; (b)examine the interactions and impacts between coastal andopen ocean regions; and (c) study the processes andfactors that control the accuracy of the reconstruction ofthe coastal ocean state. The strategy implies using OSEs/OSSEs to subsample the oceanic fields in order to quantifythe errors in the reconstruction of the coastal ocean stateand its variability.
Probabilistic approaches are another emerging topicthat is expected to provide new venues in the field ofcoastal and shelf monitoring and forecasting. The varioustechniques available to date appear already quite matureto provide robust tools for helping decision makers insetting up future coastal observing systems. The COFScommunity is expected to benefit from applying probabilis-tic methods to coastal problems of interest and assess thevalue of new probabilistic product types for both scienceand selected applications. Collaboration with large scaleforecasting systems would allow estimating ProbabilityDensity Functions for forcing of nested COFS systems.
In the context of rapidly developing coastal ocean fore-casting capabilities worldwide, international coordinationis essential to exchange best practices, optimize observingsystems and evaluate common data sets, with appropriatefeedback to the providers of forcing inputs. Examples ofwell integrated systems have been discussed to showcasethe value added to global systems, over a variety ofunique requirements. The Coastal Oceans and Shelf SeasTask Team in GODAE OceanView has been fostering inter-national forums and activities that have been addressingkey challenges in observing and predicting circulationand transport in the coastal and shelf seas. Building uponthis history, it aspires to play an important role towardsthe international coordination of science in support ofcoastal ocean forecasting.
AcknowledgmentsV. Kourafalou acknowledges support from NOAA(NA13OAR4830224 and NA11NOS4780045) and thanksH. Kang (UM/RSMAS) and P. Hogan (NRL-SSC) for SouthFlorida model outputs and N. Shay (UM/RSMAS) for WERAdata. M. Herzfeld is thankful to the eReefs marine modelingteam and partners (CSIRO: Commonwealth Industrial and Scien-tific Research Organization; SIEF: Science and Industry Endow-ment Fund; AIMS: Australian Institute of Marine Science). Thesystem CGOFS_ECS (material contributed by X. Zhu) is sup-ported by the National Natural Science Foundation of China (con-tract #41222038). E. V. Stanev acknowledges the input fromCOSYNA, which is supported by the Helmholtz Association ofGerman Research Centers. M. Cirano (Federal UniversityBahia) had substantial contribution in updating the Systems Infor-mation Table that provided input for the tabulated examples.
Journal of Operational Oceanography s143
Dow
nloa
ded
by [
Uni
dad
de R
ecur
sos
de I
nfor
mac
ion
Cie
ntif
ica]
at 0
0:47
18
May
201
6
References
Abramson B, Brown J, Edwards W, Murphy A, Winklerg RL.1996. Hailfinder: A Bayesian system for forecasting severeweather. Int J Forecasting. 12(1): 57–71.
Alvarez A, Mourre B. 2012. Optimal sampling designs for aglider-mooring observing network. J Atmos Ocean Technol.29(4): 601–612.
Apel H, Thieken AH, Merz B, Blöschl G. 2006. A ProbabilisticModelling System for Assessing Flood Risks. Nat Hazards.38(1-2): 79–100.
Bennett AF. 2002. Inverse Modeling of the Ocean andAtmosphere, Cambridge, University Press, p. 234.
Bloom SC, Takacs L.L, DA Silva AM, Ledvina D. 1996. Dataassimilation using Incremental Analysis Updates. MonWeather Rev. 124: 1256–1271.
Booij N, Ris RC, Holthuijsen LH. 1999. A third-generation wavemodel for coastal regions: 1. Model description and vali-dation. J Geophys Res. 104(C4): 7649–7666.
Bouffard J, Pascual A, Ruiz S, Faugère Y, Tintoré J. 2010. Coastaland mesoscale dynamics characterization using altimetry andgliders: A case study in the Balearic Sea. J Geophy Res. 115(C10), doi:10.1029/2009JC006087.
Brodie J, Waterhouse J. 2012. A critical review of environmentalmanagement of the ‘not so Great’. Barrier Reef Est, Coastaland Shelf Sci. 104-105: 1–22.
Carton JA, Giese BS. 2008. A Reanalysis of Ocean Climate UsingSimple Ocean Data Assimilation (SODA). MonWeather Rev.136: 2999–3017, doi:10.1175/2007MWR1978.1.
Chassignet EP, Hurlburt HE, Smedstad OM, Halliwell GR, HoganPJ, Wallcraft AJ, Baraille R, Bleck R. 2007. Ocean predictionwith the Hybrid Coordinate Ocean Model (HYCOM).J Marine Syst. 65: 60–83.
Chen CS, Liu H, Breadsley RC. 2003. An unstructured, finite-volume, three-dimensional, primitive equation ocean model:application to coastal ocean and estuaries. J Atmos OceanTechnol. 20: 159–186.
Chen CS, Malanotte-Rizzoli P, Wei J, Beardsley RC, Lai Z, Xue P,Lyu S, Xu Q, Qi J, Cowles GW. 2009. Application and com-parison of Kalman filters for coastal ocean problems: Anexperiment with FVCOM. J Geophy Res. 114(C05011),doi:10.1029/2007JC004548.
Cipollini P, Benveniste J, Miller L, Picot N, Scharroo R, Strub T,Vandemark D, Vignudelli S, Zoffoli S, Andersen O et al.2012. Conquering the coastal zone: a new frontier for satellitealtimetry. Proceedings of the “20 Years of Progress in RadarAltimetry” Symposium, Venice, Italy (Eds. Benveniste J,Morrow R), 9/2012, ESA Special Publication SP-710 (CD-ROM), ESA Publications Division, European SpaceAgency, Noordwijk, The Netherlands.
Devaliere EM, Hanson JL. 2009. IMEDS User Guide, IMEDSUser Guidehttp://www.frf.usace.army.mil/morphos/imeds/IMEDS_Help.htm.
Diaz H, Folland C, Manabe T, Parker D, Reynolds R, Woodruff S.2002. Workshop on advances in the use of historical marineclimate data. W Meteoro Org (WMO) Bulletin 51(4): 377–380.
Dobricic S, Pinardi N, Testor P, Send U. 2010. Impact of dataassimilation of glider observations in the Ionian Sea(Eastern Mediterranean). Dynam Atmos Oceans. 50(1):78–92.
Escudier R, Bouffard J, Pascual A, Poulain PM, Pujol MI. 2013.Improvement of coastal and mesoscale observation from
space: Application to the northwestern Mediterranean Sea.Geophys Res Lett. 40: 2148–2153.
European Commission. 2010. Marine Knowledge 2020 marinedata and observation for smart and sustainable growth.COM (2010) 461: 1–13.
European Commission. 2012. Green Paper-Marine Knowledge2020—from seabed mapping to ocean forecasting. p. 23ISBN 978-92-79-25350-8. doi:10.2771/4154.
European Commission. 2013. Towards European IntegratedOcean Observation. Expert Group on Marine ResearchInfrastructures Final Report. P. 96. ISBN 978-9279-27319-3, doi:10.2777/29343.
Evensen G. 1994. Sequential data assimilation with a nonlinearquasi-geostrophic model using Monte Carlo methods to fore-cast error statistics. J Geophys Res. 99(C5): 10 143–10 162.
Feyen J, Funakoshi Y, Van der Westhuysen AJ, Earle S, CarusoMagee C, Tolman HL, Aikman F. 2013. Establishing aCommunity-Based Extratropical Storm Surge and TideModel for NOAA’s Operational Forecasts for the Atlanticand Gulf Coasts. Proc. 93rd AMS Annual Meeting, Austin,TX.
Frolov S, Baptista A, Wilkin M. 2008. Optimizing fixed observa-tional assets in a coastal observatory. Cont Shelf Res. 28(19):2644–2658.
Glenn S, Schofield O. 2009. Growing a distributed oceanobservatory: our view from the COOL room. Oceanography22: 128–145.
Halliwell Jr GR, Shay LK, Brewster JK, Teague WJ. 2011.Evaluation and Sensitivity Analysis of an Ocean ModelResponse to Hurricane Ivan. Mont Weather Rev. 139(3):100802101247096, doi:10.1175/2010MWR3104.1.
Halliwell GR Jr, Srinivasan A, Kourafalou V, Yang H, Willey D,Le Hénaff M, Atlas R. 2014. Rigorous Evaluation of aFraternal Twin Ocean OSSE System for the Open Gulf ofMexico. J Atmos Ocean Technol. 31(1): 105–130.
Herzfeld M, Andrewartha J. 2012. A simple, stable and accurateDirichlet open boundary condition for ocean modellingdownscaling. Ocean Model. 43–44: 1–21.
Herzfeld M, Schmidt M, Griffies S, Liang Z. 2011. Realistic testcases for limited area ocean modelling. Ocean Modelling. 37:1–34.
Herzfeld M. 2009. The role of numerical implementation on openboundary behaviour in limited area ocean models. OceanModel 27: 18–32, doi:10.1016/j.ocemod.2008.10.008.
Hurlburt HE, Hogan PJ. 2000. Impact of 1/8 to 1/64 resolution onGulf Stream model–data comparisons in basin-scale subtropi-cal Atlantic Ocean models. Dynam Atmos Oceans. 32(3):283–329.
Jacob R, Larson J, Ong E. 2005. M × N communication and par-allel interpolation in Community Climate System ModelVersion 3 using the model coupling toolkit. InternationalJournal of High P Comp Appl. 19(3): 293–307.
Jelesnianski CP, Chen J, Shaffer WA. 1992. SLOSH: Sea, lake,and overland surges from hurricanes. NOAA TechnicalReport NWS 48, National Oceanic and AtmosphericAdministration, U.S. Department of Commerce, p. 71.
Jolliffe IT, Stephenson DB. 2011. Forecast Verification: APractitioner’s Guide in Atmospheric Science, 2nd Edition.Wiley, ed., 292 p.
Kourafalou VH, Kang H. 2012. Florida Current meandering andevolution of cyclonic eddies along the Florida Keys ReefTract: are they inter-connected? J Geophy Res. 117(C05028), doi:10.1029/2011JC007383.
Kourafalou VH, Androulidakis YS. 2013. Influence ofMississippi induced circulation on the Deepwater Horizon
Kourafalou VH, De Mey P, Staneva J, Ayoub N, Barth A, Chao Y,Cirano M, Fiechter J, Herzfeld M, Kurapov A, Moore AM,Oddo P, Pullen J, Van Der Westhuysen A, Weisberg R.2015. Coastal Ocean Forecasting: science drivers and userbenefits. J Oper Oceanogr. doi:10.1080/1755876X.2015.1022348.
Kurapov AL, Egbert GD, Allen JS, Miller RN. 2009. Representer-based analyses in the coastal upwelling system. DynamAtmos Oceans. 48: 198–218.
Kurapov AL, Foley D, Strub PT, Egbert GD, Allen JS. 2011.Variational assimilation of satellite observations in a coastalocean model off Oregon. J Geophy Res. 116(C5), doi:10.1029/2010JC006909.
Larson J, Jacob R, Ong E. 2004. The Model CouplingToolkit: A New Fortran90 Toolkit for BuildingMultiphysics Parallel Coupled Models. Preprint ANL/MCSP1208-1204. Mathematics and Computer ScienceDivision, A Nat Lab, 25.
Lazure P, Dumas F. 2008. An external-internal mode coupling fora 3D hydrodynamical model for applications at regional scale(MARS). Adv Water Resour. 31: 233–250.
Le Hénaff M, De Mey P, Marsaleix P. 2009. Assessment of obser-vational networks with the Representer Matrix Spectramethod—application to a 3D coastal model of the Bay ofBiscay. Ocean Dynam. 59(1): 3–20.
Le Hénaff M, De Mey P, Mourre B, Le Traon P-Y. 2008.Contribution of a Wide-Swath Altimeter in a shelf seas assim-ilation system: impact of the satellite roll errors. J AtmoOcean Technol. 25: 2133–2144.
Leblond E, Lazure P, Laurans M, Rioual C, Woerther P, QuemenerL, Berthou P. 2010. RECOPESCA: A new example of parti-cipative approach to collect in-situ environmental and fish-eries data. Joint Coriolis-Mercator Ocean QuarterlyNewsletter 37: 40–55.
Luettich Jr R, Westerink JJ, Scheffner NW. 1992. ADCIRC: AnAdvanced Three-Dimensional Circulation Model ofShelves, Coasts, and Estuaries, Report 1: theory and method-ology of ADCIRC-2DDI and ADCIRC-3DL. U.S.Department of the Army, Technical Report DRP-92-6.
Luettich Jr RA, Wright LD, Signell R, Friedrichs C, Friedrichs M,Harding J, Fennel K, Howlett E, Graves S, Smith E, Crane G,Baltes R. 2013. Introduction to special section on The U.S.IOOS Coastal and Ocean Modeling Testbed. J Geophy Res.118(C12): 6319–6328.
Moore AM, Arango HG, Broquet G, Edward C, Veneziani M,Powell B, Foley D, Doyle JD, Costa D, Robinson P. 2011b.The Regional Ocean Modeling System (ROMS) 4-dimen-sional variational data assimilations systems: Part III—Observation impact and observation sensitivity in theCalifornia Current. Prog Oceanography 91: 74–94.
Moore AM, Arango H, Broquet G, Powell B, Weaver AT, Zavala-Garay J. 2011a. The Regional Ocean Modeling System(ROMS) 4-dimensional variational data assimilationssystems, Part I - System overview and formulation. ProgOceanography. 91: 34–49.
Mourre B, Alvarez A. 2012. Benefit assessment of glider adaptivesampling in the Ligurian Sea. Deep Sea Research Part I 68:68–78.
Mourre B, De Mey P, Ménard Y, Lyard F, Le Provost C. 2006.Relative performance of future altimeter systems and tidegauges in constraining a model of North Sea high-frequencybarotropic dynamics. Ocean Dynam. 56: 473–486.
Oke PR, Larnicol G, Jones EM, Kourafalou VH, Sperrevik AK,Carse F, Tanajura CAS, Mourre B, Tonani M, BrassingtonGB et al. 2015. Assessing the impact of observations onocean forecasts and reanalyses: Part 2: Regional applications.J Oper Oceanogr. 7(3) (Special)
Onogi K, Tsutsui J, Koide H, Sakamoto M, Kobayashi S,Hatsushika H, Matsumoto T, Yamazaki N, Kamahori H,Takahashi K et al. 2007. The JRA-25 Reanalysis.J Meteorol Soc Jpn. 85: 369–432.
Parks AB, Shay LK, Johns WE, Martinez-Pedraja J, Gurgel K-W.2009. HF radar observations of small-scale surface currentvariability in the Straits of Florida. J Geophy Res. 114(C08002), doi:10.1029/2008JC005025.
Pascual A, Bouffard J, Ruiz S, Nardelli BB, Vidal-Vijande E,Escudier R, Sayol JM, Orfila A. 2013. Recent improvementsin mesoscale characterization of the western Mediterraneansea: Synergy between satellite altimetry and other observa-tional approaches. Sci Mar. 77: 19–36.
Pascual A, Ruiz S, Tintoré J. 2010. Combining new and conven-tional sensors to study the Balearic Current. Sea Technol. 51(7): 32–36.
Pelikan M. 2005. Hierarchical Bayesian optimization algorithms:towards a new generation of evolutionary algorithms.Springer ed., 171pp.
Proctor R, Howarth M. 2008. Coastal observatories and oper-ational oceanography: a European perspective. Mar TechnolSoc J. 42(3): 10–13.
Purvis MJ, Bates PD, Hayes CM. 2008. A probabilistic method-ology to estimate future coastal flood risk due to sea levelrise. Coast Eng. 55(12): 1062–1073.
Raicich F, Rampazzo A. 2003. Observing system simulation exper-iment for the assessment of temperature sampling strategies inthe Mediterranean Sea. Annales Geophy. 21: 151–165.
Rappaport EN, Jiing J-G, Landsea CW, Murillo ST, Franklin JL.2012. The Joint Hurricane Test Bed: Its First Decadeof Tropical Cyclone Research-To-Operations ActivitiesReviewed. Bul Am Meteorol Soc. 93: 371–380. http://dx.doi.org/10.1175/BAMS-D-11-00037.1.
Rixen M, Ferreira-Coelho E, Signell R. 2008. Surface drift predic-tion in the Adriatic Sea using hyper-ensemble statistics onatmospheric, ocean and wave models: Uncertainties and prob-ability distribution areas. J Mar Syst. 69(1-2): 86–98.
Robert CP, Casella G. 2004. Monte Carlo statistical methods, 2ndedition. Springer, ed., 645pp.
Robinson AR, Brink KH, Ducklow HW, Jahnke RA, RotschildBJ. 2004. Robinson A, Brink KInterdisciplinary multiscalecoastal dynamical processes and interactions. In: The Sea13: 3–35.
Ruiz S, Pascual A, Garau B, Faugère Y, Alvarez A, Tintoré J.2009. Mesoscale dynamics of the Balearic Front, integratingglider, ship and satellite data. J Mar Syst. 78(1): S3–S16.
Saha S, Moorthi S, Pan HL and co-authors. 2010. The NCEPClimate Forecast System Reanalysis. Bul Am MeteoroloSoc. 91: 1015–1057, doi:10.1175/2010BAMS3001.1.
Schiller A, Herzfeld M, Brinkman R, Stuart G. 2014. Monitoring,Predicting and Managing one of the Seven Natural Wondersof the World. Bul Am Meteorol Soc. 95(1): 23–30, http://dx.doi.org/10.1175/BAMS-D-12-00202.1.
Schiller A, Bell M, Brassington G, Brasseur P, Barciela R, DeMey P, Dombrowsky E, Gehle M, Hernandez F, KourafalouV, Larnicol G, Le Traon P-Y, Martin M, Oke P, Smith G,Smith N, Tolman H, Wilmer-Becker K. 2015. Synthesis ofNew Scientific Challenges for GODAE OceanView. J OperOceanogr. forthcoming.
Schiller RV, Kourafalou VH. 2010. Modelling river plumedynamics with the Hybrid Coordinate Ocean Model. OceanModel. 33(1–2): 101–117, doi:10.1016/j.ocemod.2009.12.005.
Schiller RV, Kourafalou VH, Hogan P, Walker ND. 2011. Thedynamics of the Mississippi River plume: Impact of topogra-phy, wind and offshore forcing on the fate of plume waters. JGeophy Res. 116(C6): 1978–2012.
Schulz-Stellenfleth J, Stanev EV. 2010. Statistical assessment ofocean observing networks–A study of water level measure-ments in the German Bight. Ocean Model. 33(3–4): 270–282.
Shchepetkin AF, McWilliams JC. 2005. The regional oceanicmodeling system (ROMS): a split-explicit, free-surface, topo-graphy-following-coordinate oceanic model. Ocean Model. 9:347–404, doi:10.1016/j.ocemod.2004.08.002.
Skamarock W, Klemp J, Dudhia J, Gill D, Barker DM, Wang W,Powers JG. 2005. A description of the Advanced ResearchWRF Version 2. NCAR Technical Note TN-468STR, 88pp.
Sponaugle S, Lee TN, Kourafalou VH, Pinkard D. 2005. FloridaCurrent frontal eddies and the settlement of coral reef fishes.Limnol Oceanogr. 50: 1033–1048.
Stanev E, Schulz-Stellenfleth J, Staneva J, Grayek S, Seemann J,Petersen W. 2011. Coastal observing and forecasting systemfor the German Bight-Estimates of hydro-physical states.Ocean Sci. 7: 569–583.
Taylor A, Glahn B. 2008. Probabilistic guidance for hurricane stormsurge. Proc. 88th AMS Annual Meeting, New Orleans, LA.
Tintoré J, Vizoso G, Casas B, Heslop E, Pascual A, Orfila A, RuizS, Martínez-Ledesma M, Torner M, Cusí S et al. 2013.SOCIB: the Balearic Islands Observing and ForecastingSystem responding to science, technology and societyneeds. Mar Technol Soc. J. 47(1): 101–117 http://doi.org/10.4031/MTSJ.47.1.10.
Tolman HL, Balasubramaniyan B, Burroughs LD, Chalikov DV,Chao YY, Chen HS, Gerald VM. 2002. Development andimplementation of wind generated ocean surface wavemodels at NCEP. Weat For. 17: 311–333.
Tsujino H, Motoi T, Ishikawa I, Hirabara M, Nakano H,Yamanaka G, Yasuda T, Ishizaki H. 2011. Referencemanual for the Meteorological Research Institute communityocean model (MRI.COM) version 3. Tech. Rep., 59,Meteorological Research Institute, Tsukuba, Japan.
Usui N, Ishizaki S, Fujii Y, Tsujino H, Yasuda T, Kamachi M.2006. Meteorological Research Institute multivariate ocean
variational estimation (MOVE) system: Some early results.Adv Space Res. 37(4): 806–822.
Van der Westhuysen AJ, Padilla R, Santos P, Gibbs A, Gaer D,Nicolini T, Tjaden S, Devaliere EM, Tolman HL. 2013.Development and validation of the Nearshore Wave PredictionSystem. Proc. 93rd AMS Annual Meeting, Austin, TX.
Vandenbulcke L, Beckers J-M, Lenartz F, Barth A, Poulain P-M,Aidonidis M, Meyrat J, Ardhuin F, Tonani M, Fratianni C,Torrisi L, Pallela D, Chiggiato J, Tudor M, Book JW,Martin P, Peggion G, Rixen M. 2009. Super ensemble tech-niques: application to surface drifter prediction. ProgOceanogr. 82(3): 149–167.
Warner JC, Armstrong B, He R, Zambon JB. 2010. Developmentof a coupled ocean–atmosphere–wave–sediment transport(COAWST) modeling system. Ocean Model. 35(3): 230–244.
Wolanski E, Hamner WM. 1988. Topographically controlledfronts in the ocean and their biological influence. Science241: 177–181.
Wolanski E, Asaeda T, Tanaka A, Deleersnijder E. 1996. Three-dimensional wakes in the field, laboratory experiments andnumerical models. Cont Shelf Res. 16(11): 1437–1452.
Wolanski E, Brinkmann R, Spagnol S, McAllister F, SkirvingCSW, Deleersnijder E. 2003a. Merging scales in models ofwater circulation: perspectives from the Great Barrier Reef.In: Lakhan V (Ed.), Advances in Coastal Modeling.Elsevier. pp. 411–429.
Wolanski E, Marshall K, Spagnol S. 2003b. Nepheloid layerdynamics in coastal waters of the Great Barrier Reef,Australia. J Coastal Res. 19: 748–752.
Yildirim B, Chryssotomidis C, Karniadakis GE. 2009. Efficientsensor placement for ocean measurement using low-dimen-sional concepts. Ocean Model. 27: 160–173.
Yu P, Kurapov AL, Egbert GD, Allen JS, Kosro PM. 2012.Variational assimilation of HF radar surface currents in acoastal ocean model off Oregon. Ocean Model. 49-50: 86–104.
Zhang WG, Wilkin JL, Levin JC. 2010. Towards an integratedobservation and modeling system in the New York Bightusing variational methods. Part II. Representer-based observ-ing system evaluation. Ocean Model. 35: 134–145.
Zhang Y, Baptista AM. 2008. SELFE: a semi-implicit Eulerian–Lagrangian finite-element model for cross-scale ocean circu-lation. Ocean Model. 21(3): 71–96.
